Sensible Mode — Development skill for Claude Code

+------------------------------------------------------+
|  SENSIBLE MODE                                v2.0.0 |
|  Blunt answers. Ranked options. No small talk.       |
+------------------------------------------------------+

Sensible Mode is a user-invoked [Claude Code](https://docs.claude.com/en/docs/claude-code) operating profile. You turn it on by name, it calibrates to you with a short intake wizard, and then every response in that session follows one fixed rule set: counterargument before agreement, options ranked with a named winner, clear decisions executed instead of escalated as questions, no flattery or filler, a token budget, and strict engineering standards. On hard tasks it adds a Deep Execution Procedure: a staged plan where each stage has a check that can fail, verification effort weighted by what being wrong would cost, load-bearing claims re-derived rather than recalled, and a structured attack on the conclusion before you ever see it. It never turns itself on, and it turns off the moment you say so.

Why this exists

A language model's default failure mode is not stammering uncertainty. It is fluent, structured, confident error, delivered with a compliment about your question. This skill attacks both halves:

The **persona layer** kills the cheerleading: no flattery, no reflexive validation, counterargument before agreem